A Consistent Quantum Ontology Robert B. Griffiths Physics Department Carnegie-Mellon University Version of 19 May 2011 Abstract The (consistent or decoherent) histories interpretation provides a consistent realistic ontology for quantum mechanics, based on two main ideas. First, a logic (system of reasoning) is employed which is compatible with the Hilbert-space structure of quantum mechanics as understood by von Neumann: quantum properties and their negations correspond to subspaces and their orthogonal complements. It employs a special (single framework) syntactical rule to construct meaningful quantum expressions, quite different from the quantum logic of Birkhoff and von Neumann. Second, quantum time development is treated as an inherently stochastic process under all circumstances, not just when measurements take place. The time-dependent Schrödinger equation provides probabilities, not a deterministic time development of the world. The resulting interpretive framework has no measurement problem and can be used to analyze in quantum terms what is going on before, after, and during physical preparation and measurement processes. In particular, appropriate measurements can reveal quantum properties possessed by the measured system before the measurement took place. There are no mysterious superluminal influences: quantum systems satisfy an appropriate form of Einstein locality. This ontology provides a satisfactory foundation for quantum information theory, since it supplies definite answers as to what the information is about. The formalism of classical (Shannon) information theory applies without change in suitable quantum contexts, and this suggests the way in which quantum information theory extends beyond its classical counterpart. Contents 1 Introduction 2 2 System at One Time 5 2.1 Phase Space and Hilbert Space................................... 6 2.2 Frameworks.............................................. 8 2.3 The Single Framework Rule..................................... 9 3 Time Development 11 3.1 Quantum dynamics: histories.................................... 11 3.2 Unitary families and the uniwave.................................. 13 3.3 Born rule............................................... 13 3.4 Multiple times: consistency..................................... 14 3.5 Which history occurred?....................................... 15 4 Classical Limit 16 4.1 Quasiclassical frameworks...................................... 16 4.2 Persistence of quasiclassical behavior................................ 17 5 Preparation and Measurement 18 5.1 Measurement model......................................... 18 5.2 Preparation model.......................................... 20 5.3 POVMs................................................ 21 6 Quantum Locality 22 6.1 Genuine nonlocality......................................... 22 6.2 Spurious nonlocality......................................... 23 6.3 Einstein locality........................................... 24 1
7 Quantum Information 25 7.1 Histories approach.......................................... 25 7.2 Sources and information....................................... 25 8 Conclusions 26 8.1 Summary............................................... 26 8.2 The new logic............................................. 27 8.3 Indeterministic dynamics...................................... 28 8.4 Open issues.............................................. 29 8.4.1 Entangled histories...................................... 29 8.4.2 Sufficiency of the language................................. 30 8.4.3 Thermodynamic irreversibility............................... 30 8.4.4 Epistemology......................................... 31 9 Acknowledgments 31 1 Introduction Scientific advances can significantly change our view what the world is like, and one of the tasks of the philosophy of science is to take successful theories and tease out of them their broader implications for the nature of reality. Quantum mechanics, one of the most significant advances of twentieth century physics, is an obvious candidate for this task, but up till now efforts to understand its broader implications have been less successful than might have been hoped. The interpretation of quantum theory found in textbooks, which comes as close as anything to defining standard quantum mechanics, is widely regarded as quite unsatisfactory. Among philosophers of science this opinion is almost universal, and among practicing physicists it is widespread. It is but a slight exaggeration to say that the only physicists who are content with quantum theory as found in current textbooks are those who have never given the matter much thought, or at least have never had to teach the introductory course to questioning students who have not yet learned to shut up and calculate! On all sides it is acknowledged that the major difficulty is the quantum measurement problem. Significantly, it occupies the very last chapter of von Neumann s 1932 Mathematical Foundations of Quantum Mechanics [1], and forms what many regard as the least satisfactory feature of this monumental work, the great-grandfather of current textbooks. The difficulties in the way of using measurement as a fundamental component of quantum theory were summed up by Wigner in 1963 [2], and confirmed by much later work; see, e.g., the careful analysis by Mittelstaedt [3]. A more recent review by Wallace [4] testifies both to the continuing centrality of the measurement problem for the philosophy of quantum mechanics, and to the continued lack of progress in resolving it; all that has changed is the number and variety of unsatisfactory solutions. Actually there are two distinct measurement problems. The first measurement problem, widely studied in quantum foundations, comes about because if the time development of the measurement apparatus (and its environment, etc.) is treated quantum mechanically by integrating Schrödinger s equation, the result will typically be a macroscopic quantum superposition or Schrödinger cat in which the apparatus pointer we shall continue to use this outdated but picturesque language does not have a definite position, so the experiment has no definite outcome. Contrary to the belief of experimental physicists. If this first problem can be solved by getting the wiggling pointer to collapse down into some particular direction, one arrives at the second measurement problem: how is the pointer position related to the earlier microscopic situation which the apparatus was designed to measure, and which the experimental physicist believes actually caused the pointer to point this way and not that way? When one hears experimental particle physicists give talks, it sounds as if they believe their detectors are triggered by the passage of real particles zipping along at enormous speed and producing electrical pulses by ionizing the matter through which they pass. No mention of the sudden collapse of (the modern electronic counterpart of) the apparatus pointer at the end of the measurement process. Instead, these physicists believe not only that the apparatus is in a well-defined macroscopic state each bit in the memory is either 0 or 1 after the measurement, but in addition that this outcome is well correlated with a prior state of affairs: one can be quite confident that a negative muon was moving along some specified path at a particular moment in time. Have they forgotten what they learned in their first course in quantum theory? 2
It this connection note that laboratory measurements in particle physics are, in the vast majority of cases, not appropriately modeled by the scheme proposed by von Neumann in which a seemingly arbitrary wave function collapse leads to a correlation between the measurement outcome and a property of the measured system at a time after the measurement has taken place. In a typical particle physics experiment the particle will either have disintegrated, escaped from the apparatus, or been absorbed long before the measuring apparatus has registered its behavior, and in any case the experimentalist is interested in what was going on before detection rather than afterwards. The absence of a satisfactory solution to this second measurement problem has led to the development of a certain quantum orthodoxy which affirms that the only task of quantum theory is to relate the outcome of a macroscopic measurement to an earlier and equally macroscopic (i.e., describable in ordinary language) preparation. 1 The quantum wave function ψ(t) associated with the prepared system, regarded as isolated from its environment until it interacts with the measuring apparatus (the uniwave in the terminology of Sec. 3.2 below) does not have any ontological reference; it is simply a mathematical device for calculating the probabilities of measurement outcomes in terms of the earlier preparation procedure. Stated slightly differently, it provides an abstract representation of the preparation process, or at least as much of, or that part of, the preparation process which is needed to calculate probabilities of outcomes of future measurements. In this approach preparation and measurement can be thought of as part of the real world, but what happens in between occurs inside a black box which cannot be opened for further analysis and is completely outside the purview of quantum theory although some future theory, as yet unknown, might allow a more precise description. One can be sympathetic with strict orthodoxy in that it is intended to keep the unwary out of trouble. Careless thinkers who dare open the black box will fall into the quantum foundations Swamp, where they risk being consumed by the Great Smoky Dragon, driven insane by the Paradoxes, or allured by the siren call of Passion at a Distance into subservience to Nonlocal Influences. Young scientists and philosophers who do not heed the admonitions of their elders will, like the children in one of Grimms fairy tales, have to learn the truth by bitter experience. The measurement problems and the associated lack of a clear conceptual foundation for quantum theory have not only been a stumbling block in quantum foundations work. They have also slowed down, though fortunately not stopped, mainstream physics research. In older fields such as scattering theory, the pioneers spent a significant amount of time working through conceptual issues. But once the accepted formulas are in place and yield results consistent with experiment, their intellectual descendents have the luxury of calculating without having to rethink the issues which confused their predecessors. In fields in which quantum techniques are applied in fresh ways to new problems, such as quantum information (the technical specialty of the author of this paper), conceptual issues that have not been resolved give rise to confusion and wasted time. Both students and researchers would benefit from having the rather formal approach to measurements found in, e.g., Nielsen and Chuang [6], to mention one of the best known books on the subject, replaced by something which is clearer and more closely tied to the physical intuition needed to guide good research, even in a field heavily larded with mathematical formalism. Black boxes can be a useful approach to a problem, but can also stand in the way of a good physical understanding. At one time it was optimistically supposed that quantum information would provide a new key to resolving the problems of quantum foundations [7, 8]. However, later developments have not confirmed this earlier optimism, and Timpson s 2008 review [9] and his more recent [10] provide a clear indication of where the trouble lies. Bell s question, Information about what? [11], has not been answered. And why not? Timpson realizes that quantum information cannot simply be about outcomes of measurements (assuming the first measurement problem has been solved), for this fails to connect these outcomes with properties of the system being measured. And he rejects the idea that measurements can reveal microscopic quantum properties, for this leads, in his opinion, to hidden variables and all the insuperable difficulties associated therewith. Clearly the problem is a lack of a suitable quantum ontology, something which quantum information could be about. (The possibility that quantum information could be about nothing at all is also discussed by Timpson [10] under the heading of immaterialism, which he does not find satisfactory. For comments on his own proposal in [9] for defining quantum information see Sec. 7.2.) The thesis of this paper is that both measurement problems can be, and in fact have been, resolved: the motion of the pointer stilled and the black box opened, by a consistent quantum ontology that builds upon two central ideas. The first is a system of logic that addresses the question of how to reason about a quantum system described mathematically by a Hilbert space. The second is a system of stochastic or 1 See, for example, [5]. 3
random dynamics that applies to all quantum dynamical processes, not just measurements. These ideas were brought together for the first time in the author s consistent histories interpretation [12]. Subsequently they were developed by Omnès, whose work has appeared in numerous papers and two books [13, 14], and further developed, to some extent independently, by Gell-Mann and Hartle [15] using the name decoherent histories. The differences between decoherent and consistent histories are not sufficient (in the author s opinion) to merit separate discussions, so the single term histories will be employed below; anyone who disagrees is welcome to prepend consistent wherever desired. The most complete discussion of histories ideas currently available is the author s [16], hereafter referred to as CQT; for more compact treatments see [17,18] and the first part of [19]. 2 As is often the case for new ideas, the histories approach was subject so serious criticisms by (among others) d Espagnat [20, 21], Dowker and Kent [22], Kent [23 25], and Bassi and Ghirardi [26, 27], during the decade and a half that followed the original publications. Responses were published in [28 31]; in some cases a further reply to the response will be found immediately after the reply. While these criticisms were (in the author s opinion) largely based upon misunderstandings of the histories program, they had the good effect of leading to a better and clearer formulation of its basic concepts. Vigorous scientific debate is often beneficial in this way, though it becomes ineffective if criticism are cited and responses thereto ignored. To the extent that a lack of clarity on the part of the advocates of the histories approach may be one source of misunderstanding, it is hoped that the present paper provides some remedy. An approach to quantum foundations that not only resolves both measurement problems but in addition a whole host of quantum paradoxes six chapters are devoted to such paradoxes in CQT and is consistent with special relativity (no mysterious nonlocal influences) deserves careful attention. The approach to quantum ontology presented here starts by assuming that classical mechanics, with its phase space and Hamiltonian equations of motion, embodies much of what one might hope would be true of quantum mechanics: a clean mathematical structure, an intuitive but reasonably plausible way to associate the mathematics with (what realists believe to be) the reality out there, and a system of interpretation in which human beings can seen as part of, but not an essential component in, the physical world when described in physical terms. Of course, quantum mechanics must be different from classical mechanics in some important way, as otherwise we quantum physicists have been wasting our time. But whatever differences there are at the microscopic level, the older classical ontology should be seen to emerge from, or at least be consistent with, the more fundamental quantum perspective. Focusing on the changes needed when moving from the classical to the quantum world has two fundamental advantages. First, classical mechanics has been around for a long time, and we can claim to understand it, and the associated realistic ontology, reasonably well. So our journey begins at a well-defined location, rather than with complete ignorance. Second, this route avoids getting entangled in various philosophical issues, such as the ultimate (un)reliability of human knowledge, which beset both classical and quantum ontology. Putting them aside will allow a focus on a few central issues, and the author to stay within areas where he can claim some competence. The quantum ontology presented below has the following features and consequences. First, it has no measurement problem; equivalently, it resolves both measurement problems. Second, the results are fully consistent with textbook formulations of quantum theory, once one comes to see that the textbook approach provides a set of very successful and reasonably efficient calculational tools, rather than a basic conceptual understanding of the quantum world. It is now possible to see how these calculational tools arise out of a fully consistent quantum perspective. Third, quantum mechanics is a local theory in which mysterious nonlocal influences no longer play a role [32]. Thus quantum mechanics is fully consistent with special relativity [33], contrary to claims made in some quarters. Fourth, the entire world of classical physics emerges, is consistent with, quantum physics: classical mechanics in appropriate circumstances is an approximation, sometimes an excellent approximation, to the underlying and more exact quantum mechanics which encompasses all mechanical processes at whatever length scale. In circumstances where classical mechanics applies, ordinary logic suffices for discussing physics, and can be seen to be consistent with the more general mode of reasoning required in the quantum domain. Fifth, quantum mechanics is compatible with the traditional idea of an independent physical reality whose fundamental properties can be discussed without needing to make reference to human observers or human consciousness. Sixth, this ontology provides a foundation for quantum information; it supplies a specific answer to Bell s question as to what quantum information is about. 2 It is somewhat unfortunate that the discussion of consistent histories presented in [4] bears little resemblance to what is found in [16], despite the latter being listed in the bibliography. 4
All of this at what price? First, the quantum world must be understood using an appropriate form of reasoning with features which differ not only from ordinary propositional logic, but also from the quantum logic proposed by Birkhoff and von Neumann. Second, determinism must be abandoned: quantum time development is irreducibly stochastic in all circumstances, not just when measurements occur. The remainder of the paper is organized as follows. The ontology of a quantum system at a single time is the topic of Sec. 2, where it is developed in analogy with classical phase space. In particular, the ontology of Hilbert space quantum mechanics necessarily differs from classical mechanics if one follows (at least part way) von Neumann s interpretation of the basic quantum formalism. The logical problem this poses is discussed, along with the solution proposed by the author and Omnès. Quantum time development is the topic of Sec. 3. Here the fundamental idea goes back to Born [34], but is further developed: the proper use of Schrödinger s time dependent equation, unitary time development, is to compute probabilities. The notion of a family of histories, needed for a proper probabilistic framework of quantum dynamics, is introduced, along with the technique needed for assigning probabilities in a consistent way to histories inside a closed quantum system. Following Secs. 2 and 3, which form the heart of the paper, some additional topics are discussed in a more cursory manner. Any viable quantum ontology must be able to make sense of the everyday classical world of our ordinary experience, and the strategy used to do this in the histories approach is described in Sec. 4. That program is not yet complete, but nothing known at present seems to stand in the way of its full realization, once a misunderstanding of the histories approach going back to Dowker and Kent [22] has been disposed of. Preparations and measurements are the subject of Sec. 5, which indicates the essentials needed to resolve both measurement problems. Quantum locality, including the validity of what is often referred to as Einstein locality, is treated briefly in Sec. 6 (details will be found in [19,32]). Section 7 indicates in broad strokes how the ontology presented here provides a foundation for quantum information. Following a brief overall summary in Sec. 8.1, Sec. 8.2 of the concluding Sec. 8 is devoted to a discussion of the logical issues which seem to be at the center of most criticisms of the histories approach, and which need to be clearly understood, whatever conclusion the reader may eventually wish to draw. Section 8.3 contains a few additional remarks about probabilistic dynamics. Finally some open issues, two referring to the histories approach itself and two to its wider applications to problems in the philosophy of science, are mentioned briefly in Sec. 8.4 2 System at One Time p P P P x Q Q (a) (b) Figure 1: (a) Classical phase space; (b) Hilbert space 5
2.1 Phase Space and Hilbert Space Starting with classical phase space and classical dynamics, what changes are needed to arrive at the corresponding concepts for quantum theory? The phase space (position x, momentum p) of a particle moving in one dimension is shown in Fig. 1(a), while (b) is a somewhat schematic representation of a twodimensional (complex) Hilbert space, 3 the closest quantum counterpart to phase space, for the simplest of quantum systems: the spin angular momentum of a spin-half particle (a single qubit in the jargon of quantum information theory). Following von Neumann [1] we assume that a single point in classical phase space corresponds to a ray or one-dimensional subspace in the Hilbert space: all multiples of a nonzero ψ by an arbitrary (complex) number. A ray is a single line through the origin in part (b) of the figure. In addition, the counterpart of a classical property P, such as the energy is between 1.9 and 2.0 Joules, represented by some (measurable) collection of points P in the classical phase space, is a quantum property represented by a (closed) subspace of the quantum Hilbert space. The classical property can be represented by an indicator function P(γ), where γ denotes a point in the classical phase space, and P(γ) is 1 at all points where the property holds or is true, and 0 at all other points. The quantum counterpart of an indicator function is a projector, a Hermitian operator on the Hilbert space that is equal to its square, so its eigenvalues are 0 and 1. It projects onto the subspace corresponding to the quantum property; any ket in this subspace is an eigenvector with eigenvalue 1. The negation P, NOT P, of a classical property corresponds to the set-theoretic complement P c of P, with indicator I P; I is the function whose value is 1 everywhere on the phase space. Following von Neumann we assume that the negation of a quantum property is the orthogonal complement P of the corresponding subspace P, with projector I P, where I is the identity operator on the Hilbert space. Von Neumann s proposal for associating the negation of a quantum property with the complementary subspace is central to our construction of a consistent quantum ontology, and hence it is appropriate to discuss why this approach is reasonable and to be preferred to the alternative of letting the negation of a property consist of the set-theoretic complement of the ray or subspace. First, the von Neumann approach has been widely accepted and is not (so far as the author is aware) contested by physicists, though it is not always accepted in the quantum foundations community. (We are dealing with an example of the eigenvalueeigenvector link.) It is consistent with the way students are taught in textbooks to make calculations, even though the idea itself is (alas) not always included in textbook discussions. Second, it is mathematically natural in that it makes use of a central property, the inner product, of the Hilbert space, which is what distinguishes such a space from an ordinary complex linear vector space. (The inner product defines what one means by orthogonal. ) Third, the orthogonal complement of a subspace is a subspace, whereas the set-theoretic complement is not a subspace; see, e.g., Fig. 1(b). Fourth, in the case of a spin-half particle the von Neumann proposal asserts that the complement or negation of the property S z = +1/2 (in units of h) is S z = 1/2, which is to say one of these properties is the negation of the other, in accord with the Stern-Gerlach experimental result, which showed, somewhat to the surprise of the physics community at the time, that silver atoms passing through a magnetic field gradient come out in two distinct beams, not the infinite number which would have been expected for a classical spinning particle. Or might still be expected for a spin-half quantum particle, were one to assume that all the other rays in the two-dimensional Hilbert space represent alternative possibilities to S z = +1/2. However, the Stern-Gerlach result is in complete accord with von Neumann s approach: the negation of the property corresponding to a particular ray in a two-dimensional Hilbert space is the unique property defined by the orthogonal ray, and a measurement determines which of these two properties is correct in a given case. Fifth there is a sense in which all of chemistry is based on the idea that the electron, a spin-half particle, has only two possible spin states: one up and one down. Granted, students of chemistry find this confusing, since it is not clear how up and down are to be defined, though they eventually are bludgeoned into shutting up and calculating. Among the calculations which lead to the wrong answer if 2 is replaced by some other number is that of finding the entropy of a partially ionized gas, where it is important to take into account all degrees of freedom, and log 2 is the correct contribution from the electron spin. In addition, the modern theory of quantum information is consistent with the idea that a single qubit with its 3 We follow the convention of modern quantum information theory, where the term Hilbert space is not restricted to infinitedimensional spaces, but also includes any finite-dimensional complex linear vector space equipped with an inner product. By restricting the following discussion to finite-dimensional Hilbert spaces we avoid various technical issues that are not pertinent to the conceptual issues we are concerned with. 6
two-dimensional Hilbert space can contain at most one bit of information; see [33]. It is important to stress this point, for if one accepts von Neumann s negation a wide gap opens between classical phase space and quantum Hilbert space, one clearly visible in Fig. 1. Any point in classical phase space, any possible mechanical state of the system, is either inside or outside the set that defines a physical property, which is therefore either true or false. In the quantum case there are vast numbers of rays in the Hilbert space that lie neither inside the ray or, more generally in higher dimensions, the subspace corresponding to a property, nor in the complementary subspace that corresponds to its negation. For these the property seems to be neither true nor false, but somehow undefined. How are we to think about this situation? This is in a sense the central issue for any ontology that uses the quantum Hilbert space as the physicist s fundamental mathematical tool for representing the quantum world. One approach, typical of textbooks, is to ignore the problem and instead take refuge in measurements. But problems do not go away by simply being ignored, and ignoring this particular problem makes it re-emerge under a different name: the measurement problem. Before going further, let us restate the matter in the language of indicator functions and projectors introduced earlier. The product of two (classical) indicator functions P and Q is itself an indicator function for the property P AND Q, the intersection of the two sets of points P and Q in the phase spaces, Fig. 1(a). However, the product of two (quantum) projectors P and Q is itself a projector if and only if PQ = QP, that is if the projectors commute. Otherwise, when P Q is not equal to QP, neither product is a projector, and thus neither can correspond to a quantum property. A ray Q in a two-dimensional Hilbert space, Fig. 1(b), that coincides neither with a ray P nor its orthogonal complement P is represented by a projector Q that does not commute with P, and thus it is unclear how to define the conjunction of the corresponding properties. Von Neumann was not unaware of this problem, and he and Birkhoff had an idea of how to deal with it. Instead of using the product of two noncommuting projectors it seemed to them sensible to employ the set theoretic intersection of the corresponding subspaces of the Hilbert space, which is itself a subspace and thus corresponds to some property, as a sort of quantum counterpart of AND. The corresponding OR can then be associated with the direct sum of the two subspaces. When projectors for the two subspaces commute these geometrical constructions lead to spaces whose quantum projectors, PQ and P + Q PQ, coincide precisely with what one would expect based on the analogy with classical indicators. The resulting structure, known as quantum logic, obeys some but not all of the rules of ordinary propositional logic, as Birkhoff and von Neumann pointed out. If one naively goes ahead and applies the usual rules of reasoning with these definitions of AND and OR it is easy to construct a contradiction; for a simple example see Sec. 4.6 of CQT. While quantum logic has been fairly extensively studied it seems fair to say that this route has not made much if any progress in resolving the conceptual difficulties of quantum theory. It is not mentioned in most textbooks, and is given negligible space in [4]. Perhaps the difficulty is that physicists are simply not smart enough, and the resolution of quantum mysteries by this route will have to await the construction of superintelligent robots. (But if the robots succeed, will they be able to, or even interested in, explaining it to us?) The histories approach follows von Neumann in letting subspaces represent properties, with the negation of a property represented by the orthogonal complement of the subspace. But it takes a very different and much more conservative attitude than Birkhoff and von Neumann in the case of properties P and Q represented by noncommuting projectors. Since neither P Q nor QP is a projector, let us not try and talk about their conjunction. Let us adopt a language for quantum theory in which PQ makes sense and represents the conjunction of the two properties in those cases in which QP = PQ. But if PQ QP the statement P AND Q is meaningless, not in a pejorative sense but in the precise sense that this interpretation of quantum mechanics can assign it no meaning. In much the same way that in ordinary logic the proposition P Q, P AND OR Q, is meaningless even if P and Q make sense: the combination P Q has not been put together using the rules for constructing meaningful statements. Likewise, in quantum mechanics both the conjunction and also the disjunction P OR Q, must be excluded from meaningful discussion if the projectors do not commute. The two-dimensional Hilbert space of spin-half particle can serve as an illustration. Let [z + ] := z + z +, [z ] := z + z +, (1) where we shall (as in CQT) hereafter employ the [ ] notation for this type of dyad, be projectors for the properties S z = +1/2 and S z = 1/2; similarly [x + ] and [x ] are the corresponding properties for S x. The 7
product of [z + ] and [z ] is zero (i.e., the zero operator) in either order, so they commute and their conjunction is meaningful: it is the quantum property that is always false (the counterpart of the empty set in the case of a phase space). However, neither [z + ] nor [z ] commutes with either [x + ] or [x ], so the conjunction S x = +1/2 AND S z = +1/2 is meaningless. In support of the claim that it lacks meaning one can note that every ray in the spin-half Hilbert space has the interpretation S w = +1/2 for some direction in space w. Thus thus there are no spare rays available to represent S x = +1/2 AND S z = +1/2 ; there is no room in the Hilbert space for such conjunctions. It is very important to distinguish meaningless from false. In ordinary logic if a statement is false then its negation is true. But the negation of a meaningless statement is equally meaningless. Thus the statement S z = +1/2 AND S z = 1/2 is meaningful and always false, whence its negation S z = 1/2 OR S z = +1/2 is meaningful and always true, and is consistent with the Stern-Gerlach experiment. On the other hand, S x = +1/2 AND S z = 1/2 is meaningless, and its formal negation, S x = 1/2 OR S z = +1/2, is equally meaningless. The student who has learned quantum theory from the usual courses and textbooks may well go along with the idea that S x = +1/2 AND S z = +1/2 lacks meaning, since he can think of no way of measuring it (and has probably been told that it cannot be measured). However he will be less likely to go along with the equally important idea that the disjunction S x = +1/2 OR S z = +1/2 is similarly meaningless. Granted, there is no measurement which can distinguish the two, but he has a mental image of a spin-half particle in the state S z = +1/2 as a little gyroscope with its axis of spin coinciding with the z axis. Such mental images are very useful to the physicist, and perhaps indispensable, for they help organize our picture of the world in terms that are easily remembered; they provide physical intuition. But this particular mental image can be quite misleading in suggesting that when S z = +1/2 the orthogonal components of angular momentum are zero. However, this cannot be the case since, since as the student has been taught, S 2 x = S 2 y = 1/4. Now any classical picture is bound to mislead to some extent when one is trying to think about the quantum world, but in this case a slight modification is less misleading. Imagine a gyroscope whose axis is oriented at random, except one knows that the z component of angular momentum is positive rather than being negative. Among other things this modified image helps guard against the error that the spin degree of freedom of a spin-half particle can carry a large amount of information, when in fact the limit is one bit (log 2 2). 2.2 Frameworks Ordinary probability theory uses the concept of a sample space: a collection of mutually exclusive alternatives or events, one and only one of which occurs or is true in a particular realization of some process or experiment. One way of introducing probabilities in classical statistical mechanics is to imagine the phase space divided up into a collection of nonoverlapping cells, a coarse graining in which each cell represents one of the mutually exclusive alternatives one has in mind. Let P j (γ) be the indicator function for the j th cell: equal to 1 if γ lies within the cell and 0 otherwise. (We are using the superscript of P for a label, not an exponent; as the square of an indicator is equal to itself, this need not cause confusion.) Obviously the product P j P k of the indicator functions for two different cells is 0, and the sum of all the indicator functions is the identity function I(γ) equal to 1 for all γ: P j P k = δ jk P j ; P j = I. (2) Next, probabilities are assigned to the events making up a Boolean event algebra. If one coarse grains the phase space in the manner just indicated, an event algebra can be constructed in which each event is represented by the union of some of the cells in the coarse graining; equivalently, the event algebra is the collection of all indicator functions which are sums of some of the indicators in the collection {P j }, including the functions and I which are everywhere 0 and 1, respectively. The probabilities themselves can be specified by a collection {p j } of nonnegative real numbers that sum to 1, with the probability of an event E being the sum of those p j for which the corresponding indicator functions P j appear in the sum defining E. To be sure, classical statistical mechanics is usually constructed without using a coarse graining, employing the Borel sets as an event algebra, and then introducing an additive positive measure to define probabilities. There is nothing wrong with this, but for our purposes a coarse graining provides a more useful classical analog. The quantum counterpart of a classical sample space is referred to in the histories approach as a framework. It is a projective decomposition (PD) of the identity operator I: a collection {P j } of mutually orthogonal j 8
projectors which sum to the identity operator I, and thus formally satisfy exactly the same conditions (2) as a collection of classical indicators. The fact that P j P k vanishes for j k means that the corresponding quantum properties ( events is the customary term in probability theory) are mutually exclusive: if one is true the other must be false, and the fact that they sum to the identity operator I means that at least one, and therefore only one, is true or real or actual. The corresponding event algebra is the Boolean event algebra of all projectors which can be formed by taking sums of projectors in {P j } along with the 0 operator, which plays the same role as the empty set in ordinary probability theory. As long as the quantum event algebra and the sample space are related in this way, there is no harm in using the somewhat loose term framework to refer to either one, as we shall do in what follows. Two frameworks {P j } and {Q k } are compatible if all the projectors in one commute with all of those in the other: P j Q k = Q k P j for all values of j and k. Otherwise they are incompatible. One says that {P j } is a refinement {Q k } if the projectors in the PD of the latter are included in the event algebra of the former; equivalently, {Q k } is a coarsening of {P j }. Obviously two frameworks must be compatible if one is to be a refining or coarsening of the other. In addition, two compatible frameworks always possess a common refinement using the PD consisting of all the nonzero products P j Q k ; its event algebra includes the union of the event algebras of the separate frameworks which it refines. Note that a given framework {P j } may have various different refinements, and two refinements need not be compatible with each other. Therefore when discussing quantum systems one must keep track of the framework being employed in a particular argument. This is not important in classical physics, where one can either adopt the finest framework possible at the outset, or else refine it as one goes along without needing to call attention to this fact. In quantum mechanics one does not have this freedom, and carelessness can lead to paradoxes. 2.3 The Single Framework Rule A central concept of the histories approach is the single framework rule, which states that probabilistic reasoning that starts from data (observed or simply assumed) about a quantum system and leads to conclusions about the same system, typically expressed as conditional probabilities, is invalid unless it is carried out using a single framework as defined above. In particular it is not valid when it results from combining incompatible frameworks. As the single framework rule has been frequently misunderstood by critics of the histories approach, it is important to clarify what it does and does not mean. While it rules out improper combinations of descriptions, it does not prevent the physicist from employing a variety of different and possibly incompatible frameworks when constructing several distinct descriptions of a quantum system, each of which may provide some physical insight about its behavior. This is the principle of Liberty: the physicist can use whatever framework he chooses when describing a system. All properly constructed individual frameworks are equally acceptable in terms of fundamental quantum mechanics: the principle of Equality. The principle of Incompatibility forbids combining incompatible frameworks. Not all frameworks are equally useful in answering particular questions of physical interest, let us call this the principle of Utility. It is by combining these principles that the single framework rule arrives at a consistent quantum ontology adequate for understanding the quantum world in a realistic way, while at the same time resolving or avoiding (or taming ) the numerous paradoxes or inconsistencies that beset alternative approaches to quantum interpretation. The physicist s Liberty to choose different frameworks should not be thought of as in any way influencing reality. Choosing a description is choosing what to talk about, and what physicists choose to talk about has very little (direct) influence on what actually goes on in the world. Shall we discuss the location of Jupiter s center of mass, or its rate of rotation? Either is possible, and neither has the slightest influence on the behavior of Jupiter. What about a silver atom approaching a Stern-Gerlach apparatus? Shall we discuss its (approximate) location or the value of S x or the value of S z? Any one of these is possible, and the single framework rule allows location to be combined with S x, or with S z. But not with both, since it is impossible to put both S x and S z, at least when referring to a single particle at a particular time, in the same framework. In this case, no less than for Jupiter, the physicist s choice of framework has not the slightest influence on the silver atom. And there is no law of nature which singles out a framework that includes S x as somehow correct or true in distinction to the S z framework; that would contradict Equality. However, they are incompatible. They cannot be combined. What does this mean? Incompatibility in this technical sense is a feature of the quantum world with no exact classical analog: in classical physics all the operators commute. But classical analogies and disanalogies, together with applications to various specifically quantum situations can help tease out its intuitive meaning. Let us start 9
with a disanalogy. A coin can land heads or tails, two mutually exclusive possibilities: if one is true, the other must be false. There is a temptation to think of the relationship of the incompatible S x and S z frameworks in this way, and it must be resisted, for it leads to a serious misunderstanding. The properties S z = +1/2 and S z = 1/2 are analogous to heads and tails: they are mutually exclusive. If one is true the other is false, and the combination S z = +1/2 AND S z = 1/2 is meaningful and false, as discussed earlier. On the other hand the combination S x = +1/2 AND S z = 1/2 is meaningless, neither true nor false. Statements belonging to incompatible frameworks cannot be compared in any way, which is also why it is meaningless to say that one framework rather than the other is the true or correct way of describing the quantum world. Equality must be taken seriously. For a positive analogy, think of a framework as something like a coarse graining of the classical phase space as discussed earlier in Sec. 2.2. Many coarse grainings are possible and the physicist is at Liberty to choose one that is convenient for whatever purposes he has in mind. There is no correct coarse graining, though some coarse grainings may be more useful than others in discussing a particular problem. The physicist s choice of coarse graining does not, of course, have any influence on the system whose properties he is trying to model. In all these respects the choice of coarse graining is like the choice of a quantum framework. But classical coarse grainings of the same phase space can always be combined: the common refinement is constructed in an obvious way using cells formed by intersections of those taken from the two coarse grainings that are being refined. However, PDs of the same Hilbert space cannot in general be combined, so in this respect the analogy fails. However, it is still helpful in illustrating some aspects of the quantum situation, and in avoiding the misleading idea that the relationship between different quantum frameworks is one of mutual exclusivity. Similarly, choosing a framework is something like choosing an inertial reference frame in special relativity. The choice is up to the physicist, and there is no law of nature, at least no law belonging to relativity theory, that singles out one rather than another. Sometimes one choice is more convenient than another when discussing a particular problem; e.g., the reference frame in which the center of mass is at rest. The choice obviously does not have any influence upon the real world. But again there is a disanalogy: any argument worked out using one inertial frame can be worked out in another; the two descriptions can be mapped onto each other. This is not true for quantum frameworks: one must employ a framework (there may be several possibilities) in which the properties of interest can be described; they must lie in the event algebra of the corresponding PD. For a more picturesque positive analogy consider a mountain, say Mount Rainier, which can be viewed from different sides. An observer can choose to look at it from the north or from the south; there is no law of nature that singles out one perspective as the correct one. One can learn different from different viewpoints, so there might be some Utility in adopting one perspective rather than the other. But once again the analogy fails in that the north and south views can, at least in principle, be combined into a single unified description of Mount Rainier from which both views can be derived as partial descriptions. Let us call this the principle of unicity. It no longer holds in the quantum world once one assumes the Hilbert space represents properties in the manner discussed above. But how can Utility play a role in quantum physics? Again consider the case of a spin-half silver atom, and suppose it is midway in its trajectory from an apparatus where a competent experimentalist has prepared it in a state with S x = +1/2 to an apparatus, also constructed by a competent experimentalist, which will later measure S z with the pointer corresponding to S z = +1/2. What can one say about the spin of the atom midway between preparation and measurement, assuming it travels in a region free from magnetic fields that could cause the spin to precess? There is a framework which at the intermediate time includes the possibilities S x = +1/2 and 1/2; using this S x framework and the data about the preparation one can infer that S x = +1/2 with probability 1 and S x = 1/2 with probability 0. There is an alternative S z framework that at the intermediate time includes the possibilities S z = +1/2 and 1/2, and it can be used to infer from the later measurement outcome that at the intermediate time the (conditional) probabilities for S z = +1/2 and 1/2 are 1 and 0 respectively. The S x framework is useful if one is concerned with whether the preparation apparatus was functioning properly, while the S z framework is useful if one wants to discuss the proper functioning of the measuring device. 10
3 Time Development 3.1 Quantum dynamics: histories The histories approach treats the time dependence of a quantum system as a random or stochastic process, one in which the future and past states of the system at different times are not determined by the present state, but only related to it by certain probabilities, which only in very special cases are 0 and 1, corresponding to a deterministic time development. Indeed, most physicists accept that in practice quantum-mechanical time development is probabilistic and not deterministic. Take the case of spontaneous decay. Modern physics possesses tools for calculating decay rates of atoms in excited states, or unstable nuclei, and they work reasonably well. If the quantum world were deterministic one would expect to find somewhere in the theoretical formalism a prescription for predicting the (relatively) precise time of decay of, say a radioactive nucleus. This time can be measured quite precisely, better than a millisecond, for a nucleus with a half life of minutes or hours or even years. Most quantum physicists do not believe that there is some marker or clock inside the nucleus which before the decay can be used to the time of decay; there is no room for it in the Hilbert space description. 4 Hence the assumption of a probabilistic dynamics does not, in the modern context, represent much of an innovation. But doing it in a consistent way that avoids paradoxes is not altogether straightforward. The essential mathematical structures needed for introducing probabilities into quantum mechanics were introduced in Sec. 2.2: a sample space constituted by a projective decomposition (PD) of the identity operator I on the Hilbert space, and a Boolean event algebra of projectors generated from the PD in a natural way. What remains is to assign nonnegative probabilities p j summing to 1 to the elements P j of the sample space, and thereby to the subspaces (events) that constitute the event algebra in precisely the same way as in other applications of probability theory. If this can be done, one can then carry out probabilistic reasoning in the quantum domain following all of the ordinary rules of probability theory provided the sample space (and event algebra) remain fixed while various (conditional) probabilities are computed the single framework rule. Just as in other applications of probability theory there is no general rule that specifies the probabilities p j : they enter a probabilistic description of the world as parameters, and some exercise of judgment on the part of the scientist is generally necessary, as well as input data, results of experiments, etc. However, quantum theory introduces a new element not found in other applications of probability theory. It applies to the time development of a closed quantum system, by which we mean either that it is isolated, completely self-contained with no environment with which it interacts, or that its interaction with with its environment is well-enough approximated by assigning to the system itself a (possibly time-dependent) Hamiltonian, which can then be employed in Schrödinger s equation. In this case certain (conditional) probabilities relating states of the system at different times can be assigned through the Born rule and its extensions to more than two times, as discussed in Secs. 3.3 and 3.4. The unitary time evolution induced by Schrödinger s equation is used in calculating these probabilities, but only in exceptional cases, as discussed in Sec. 3.2, can it be used directly (without a probabilistic interpretation) to describe what is really going on in a closed quantum system. Before assigning probabilities to processes occurring in time, we need to construct an appropriate quantum sample space. How is this to be done? Classical physics provides a useful hint. The sample space for a coin tossed three times in a row consists of the eight possibilities HHH, HHT,... TTT, where HTT means heads on the first toss, tails on the second and third. This is just the Cartesian product of the sample space for a single toss of the coin, and is in fact identical to the sample space needed to describe three different coins all tossed at the same time. Hence in ordinary probability theory sequences of events at successive times in a particular system are formally the same thing as multiple copies of the same system considered at a single time. In quantum mechanics the mathematics for describing a compound system consisting of a collection of (distinguishable) subsystems is well know: one uses the tensor product of the Hilbert spaces. This immediately suggests that the way to construct a quantum sample space for a system at successive times is to use a tensor product of copies of its Hilbert space, as first proposed in [38]. Thus for a spin-half particle 4 A notable exception is the proponents of Bohmian mechanics, who add additional hidden variables to the Hilbert space description. See, e.g., [35]. However, in the case of radioactive decay they have no way of accessing this deterministic internal clock other than by observing the actual time of decay. Determinism is also upheld by followers of Everett, see, e.g., [36, 37], but they, too, have no way of accessing the alternative worlds (or minds or whatever) in which the nucleus decays at a time different from that observed in the laboratory. 11